PearlLeeStudio/TheArtist-MusicTransformer-ft-pop80

TheArtist Music Transformer — F1 (Pop 10K Mix, pop-leaning)

Jazz-adapted chord model with a 10,000-sequence pop rehearsal buffer. The pop-leaning endpoint of the mix-ratio sweep. Pop accuracy actually improves on the pre-fine-tune baseline; jazz reaches +8.17 points.

One of six checkpoints released alongside the paper Empirical Study of Pop and Jazz Mix Ratios for Genre-Adaptive Chord Generation (Lee, 2026). See the paper for full context and the collection overview at PearlLeeStudio/TheArtist-MusicTransformer-pop-baseline.

Demo

Watch TheArtist in action on YouTube — interactive staff editor, MIDI input, AI generation with live progress, and per-genre LoRA playback across the 13-genre vocabulary.

Model summary

Field	Value
Architecture	Music Transformer with relative positional attention
Parameters	25,661,440
Vocabulary size	351 tokens
Max sequence length	256
d_model / heads / FFN / layers	512 / 8 / 2048 / 8
Fine-tune resumed from	Phase 0 pop baseline
Best epoch	6

Training data

All 1,513 jazz training sequences (Jazz Harmony Treebank, JazzStandards, Weimar Jazz Database, JAAH) plus 10,000 pop rehearsal sequences sub-sampled with seed 42 from the Phase 0 pop training split. Pop:jazz ≈ 6.6:1 in the mix.

Fine-tune hyperparameters: peak learning rate 2 × 10⁻⁵, two-epoch warmup, ten epochs maximum with patience 5.

Evaluation (held-out per-genre test sets)

Metric	Pop test	Jazz test
Top-1 accuracy	84.60%	81.03%
Top-5 accuracy	96.96%	92.41%
Perplexity	1.78	2.31
Δ vs. Phase 0 baseline	+0.36	+8.17

This is the only run in the sweep whose pop top-1 exceeds the Phase 0 baseline. It is also the run with the most stable pop curve over training. Choose F1 when pop fluency is a hard constraint and jazz coloration is welcome but not the primary target. Generations stay rooted in commercial pop and rock harmony, with jazz substitutions appearing selectively (an occasional secondary dominant or ii-V detour inside an otherwise diatonic loop).

Out-of-distribution per-genre baseline (added 2026-05-11)

F1 alone (no LoRA), measured on each of the 11 Chordonomicon genre val splits that the per-genre LoRA adapters target. New genre tokens are resized into the embedding matrix at [GENRE:none] (i.e. unconditioned), since the F1 base never saw them during training. This is the no-LoRA reference reported alongside every lora-<genre> adapter card.

Genre	Val seq.	F1 top-1 (%)	F1 top-5 (%)	F1 val loss
hip-hop	1,402	86.51	96.27	0.6240
electronic	1,519	84.50	95.93	0.6835
rock	4,891	82.79	96.75	0.5865
folk	6,075	82.66	95.80	0.7406
funk	283	82.54	94.38	0.7878
country	6,173	82.45	96.22	0.7402
rnb/soul	955	82.09	94.12	0.8119
blues	994	81.70	94.80	0.8137
gospel	374	79.34	94.73	0.8813
bossa	1,431	78.33	93.64	0.9635
classical	37	43.54	72.82	2.8653

F1 extrapolates reasonably to rock, hip-hop, country, folk, and other commercial-adjacent genres without ever having seen their genre tokens. It struggles on classical (functional tonality outside the pop/jazz training distribution) and to a lesser extent on bossa and gospel. The matching lora-<genre> adapters lift each of these numbers; see the LoRA adapter cards under the same PearlLeeStudio/ namespace for the LoRA-vs-base Δ tables. Source: ai/results/f1_per_genre_baseline.csv.

Intended use and limitations

Recommended for chord-composition workflows targeting pop, rock, CCM, K-pop, J-pop, and modern country with optional jazz coloration. F4 (ft-pop29) is the symmetric jazz-leaning endpoint; F3 (ft-pop50) is the balanced middle.

Out of scope: melody or audio generation; genres outside pop, rock, and jazz; real-time low-latency settings.

Usage

The repo bundles the project's model.py and tokenizer.py at the repo root, so external users can load the checkpoint end-to-end without cloning anything from GitHub. snapshot_download materializes the full repo on disk; sys.path makes the bundled model.py / tokenizer.py importable.

Required dependencies: torch, huggingface_hub.

import sys
import torch
from huggingface_hub import snapshot_download

# Download the full repo (model.py, tokenizer.py, best.pt, config.json).
ckpt_dir = snapshot_download(repo_id="PearlLeeStudio/TheArtist-MusicTransformer-ft-pop80")
sys.path.insert(0, ckpt_dir)  # so the next two imports resolve

from model import MusicTransformer
from tokenizer import ChordTokenizer

tokenizer = ChordTokenizer()
ckpt = torch.load(f"{ckpt_dir}/best.pt", map_location="cpu", weights_only=False)
model = MusicTransformer(
    vocab_size=tokenizer.vocab_size,
    d_model=512, n_heads=8, d_ff=2048, n_layers=8,
    max_seq_len=256, dropout=0.0, pad_id=tokenizer.pad_id,
)
model.load_state_dict(ckpt["model_state_dict"])
model.eval()

# Prompt = ii-V-I in C major; ask for a pop-flavoured continuation.
song = {
    "key": "Cmaj", "time_signature": "4/4", "genre": "pop",
    "bars": [["Dm7", "G7"], ["Cmaj7"]],
}
prompt_ids = tokenizer.encode_sequence(song)[:-1]
ids = torch.tensor([prompt_ids])
with torch.no_grad():
    for _ in range(32):
        logits = model(ids)
        next_id = torch.multinomial(
            torch.softmax(logits[:, -1, :] / 0.8, dim=-1), 1,
        )
        ids = torch.cat([ids, next_id], dim=-1)
        if next_id.item() == tokenizer.eos_id:
            break
print(tokenizer.decode(ids[0].tolist()))

For per-genre adaptation beyond pop and jazz, see the 11 LoRA adapter repos at PearlLeeStudio — they chain on top of this base.

Per-genre real-song eval (held-out 130-song set, 2026-05)

First per-genre evaluation of ft-pop80 beyond the pop/jazz split that the original paper reports.

Eval results

Genre	n_songs	Top-1 (%)	Top-5 (%)	val_loss
pop	10	86.68	96.01	0.5734
rock	10	86.69	97.48	0.4578
jazz	10	64.96	81.16	1.8958
blues	10	81.52	93.91	0.8410
bossa	10	81.43	95.47	0.7825
classical	10	49.55	81.17	2.2389
country	10	85.89	98.44	0.5152
electronic	10	87.39	98.45	0.5072
folk	10	85.04	98.92	0.5244
funk	10	83.85	96.03	0.6811
gospel	10	79.79	96.85	0.7367
hip_hop	10	90.66	98.59	0.3957
rnb_soul	10	85.11	97.07	0.5877

On this eval set F1 peaks on hip_hop (90.66%) and struggles most on classical (49.55%). This is auxiliary signal — the 11 per-genre LoRAs (sister lora-* repos) are the recommended path for production use on the 9 non-pop, non-jazz genres. F-series cells on those genres show what the base model produces under [GENRE:none] conditioning (the model's [GENRE:X] token does not exist for the 9 new genres in the F-series vocab=351).

Eval dataset composition

130 songs total, 10 per genre × 13 genres. Drawn from the same splits/val.jsonl + splits/test.jsonl partitions every F-series model was held out from during training — no train-set leakage. Built by ai/training/build_eval_real_songs.py --seed 42 --per-genre 10 (deterministic).

Genre	n	Source(s)	Bar range	Avg duration · named
pop	10	billboard	58–116	189s · 10/10 named
rock	10	chordonomicon_rock	52–87	127s · 0/10 named
jazz	10	choco:jazz-corpus, choco:real-book, jazzstandards, jht	16–89	72s · 10/10 named
blues	10	chordonomicon_blues	24–46	93s · 0/10 named
bossa	10	chordonomicon_bossa	24–78	88s · 0/10 named
classical	10	chordonomicon_classical	11–40	60s · 10/10 named
country	10	chordonomicon_country	30–81	110s · 0/10 named
electronic	10	chordonomicon_electronic	25–84	89s · 0/10 named
folk	10	chordonomicon_folk	33–82	114s · 0/10 named
funk	10	chordonomicon_funk	30–60	92s · 0/10 named
gospel	10	chordonomicon_gospel	24–85	98s · 0/10 named
hip_hop	10	chordonomicon_hip_hop	24–81	136s · 0/10 named
rnb_soul	10	chordonomicon_rnb_soul	34–82	128s · 0/10 named

Source license summary: McGill Billboard (CC0, named pop songs), Jazz Harmony Treebank / JazzStandards / WJazzD (Public / community-redistributed, named jazz standards), Bach chorales via music21 (public domain, named pieces), Chordonomicon per-genre subsets (CC BY-NC 4.0; titles are Spotify track IDs by upstream dataset policy — progressions are real songs). See docs/EVAL.md for full breakdown.

Methodology

Teacher-forced next-token cross-entropy / top-1 / top-5 over each song's token sequence (BOS + key + time_sig + genre + bars + EOS, truncated to max_seq_len=256). Same evaluate() call as ai/results/f1_per_genre_baseline.csv, just narrowed to the curated 130-song subset. Token-level metrics; not a generation-quality eval (free-generation comparison with R1 Sethares + R2 theory RAG rerank is documented separately in ai/results/eval_report.md).

Caveats:

• classical val partition is intrinsically small (37 sequences in full eval); the 10-song subset here has even narrower confidence bands. Directional finding (LoRA helps a lot on Bach harmony) is robust, exact pp deltas are noisy.
• F-series numbers on the 9 LoRA-only genres are conditioned without genre tag (vocab=351 has no [GENRE:country] token etc.). This is the realistic "F-series alone" condition, not a controlled ablation.

Source CSV: ai/results/real_song_eval.csv (17 models × 130 songs, long format).

Training-data licenses

Dataset	License
Chordonomicon	Public (user-generated)
McGill Billboard	CC0
Jazz Harmony Treebank	Public
JazzStandards (iReal Pro)	Community redistribution
Weimar Jazz Database	ODbL
JAAH	Research-use public

Citation

Cite the original mix-ratio paper. The companion per-genre LoRA paper (chord-symbol time-series adaptation) is in preparation; its arXiv ID will be added here once posted.

@misc{lee2026chordmix,
  title         = {Empirical Study of Pop and Jazz Mix Ratios for Genre-Adaptive Chord Generation},
  author        = {Lee, Jinju},
  year          = {2026},
  eprint        = {2605.04998},
  archivePrefix = {arXiv}
}

@misc{lee2026chordtimeseries,
  title         = {How Far Can Chord-Symbol Time-Series Adaptation Carry Genre Identity?},
  author        = {Lee, Jinju},
  year          = {2026},
  note          = {arXiv preprint, ID TBD},
}